To evaluate the performance of our method across different platform and copy number calling tools. We carried out benchmark analysis in prostate cancer.
Compare the copy number calling algorithms
For WGS, we select copy number from ABSOLUTE and ACEseq. For WES, we select copy number from ABSOLUTE and ASCAT.
library(sigminer)
library(ggplot2)
library(tidyverse)
tally_SNP_ABSOLUTE <- readRDS("../data/benchmark/tally_SNP_ABSOLUTE_matrix.rds")
tally_SNP_ASCAT2 <- readRDS("../data/benchmark/tally_SNP_ASCAT2_matrix.rds")
# Compare different CNV algorithms in SNP
SNP_sample <- intersect(rownames(tally_SNP_ABSOLUTE), rownames(tally_SNP_ASCAT2))
tally_SNP_ABSOLUTE <- subset(tally_SNP_ABSOLUTE, rownames(tally_SNP_ABSOLUTE) %in% SNP_sample)
tally_SNP_ASCAT2 <- subset(tally_SNP_ASCAT2, rownames(tally_SNP_ASCAT2) %in% SNP_sample)
sim_profile.SNP <- sapply(SNP_sample, function(i) {
dt1.1 <- tally_SNP_ASCAT2[i, ]
dt2.1 <- tally_SNP_ABSOLUTE[i, ]
sigminer:::cosine(dt1.1, dt2.1)
}) %>%
subset(. > 0.4)
median(sim_profile.SNP)[1] 0.8500122hist(sim_profile.SNP,
breaks = 50, xlab = "",
main = "Copy number profile similarity from two methods(SNP)", xlim = range(0, 1)
)
abline(v = 0.8500122, col = "red", lty = 2)
We compared the signature between tools in SNP.
SNP_ABSOLUTE_SP_sigs <- readRDS("../data/benchmark/SNP_ABSOLUTE_SP_highmatch.rds")
SNP_ASCAT2_SP_sigs <- readRDS("../data/benchmark/SNP_ASCAT2_SP_highmatch.rds")
colnames(SNP_ABSOLUTE_SP_sigs$solution_list$S4$Signature.norm) <- paste0("SNP_ABSOLUTE_sigs", seq(1:4))
colnames(SNP_ASCAT2_SP_sigs$solution_list$S4$Signature.norm) <- paste0("SNP_ASCAT_sigs", seq(1:4))
sim <- get_sig_similarity(SNP_ABSOLUTE_SP_sigs$solution_list$S4, SNP_ASCAT2_SP_sigs$solution_list$S4)p <- pheatmap::pheatmap(sim$similarity, cluster_cols = F, cluster_rows = F, display_numbers = TRUE)
tally_WGS_ABSOLUTE <- readRDS("../data/benchmark/tally_WGS_ABSOLUTE_matrix.rds")
tally_WGS_aceseq <- readRDS("../data/benchmark/tally_WGS_aceseq_matrix.rds")
# Compare different CNV algorithms in WGS
WGS_sample <- intersect(rownames(tally_WGS_ABSOLUTE), rownames(tally_WGS_aceseq))
tally_WGS_ABSOLUTE <- subset(
tally_WGS_ABSOLUTE,
rownames(tally_WGS_ABSOLUTE) %in% WGS_sample
)
tally_WGS_aceseq <- subset(tally_WGS_aceseq, rownames(tally_WGS_aceseq) %in% WGS_sample)
sim_profile.WGS <- sapply(WGS_sample, function(i) {
dt1.1 <- tally_WGS_ABSOLUTE[i, ]
dt2.1 <- tally_WGS_aceseq[i, ]
sigminer:::cosine(dt1.1, dt2.1)
}) %>%
subset(. > 0.4)
hist(sim_profile.WGS,
breaks = 50, xlab = "",
main = "Copy number profile similarity from two methods(WGS)", xlim = range(0, 1)
)
abline(v = 0.9725061, col = "red", lty = 2)
We compared the signature between tools in WGS.
WGS_ABSOLUTE_SP_sigs <- readRDS("../data/benchmark/WGS_ABSOLUTE_SP_highmatch.rds")
WGS_aceseq_SP_sigs <- readRDS("../data/benchmark/wgs_aceseq_sig.rds")
colnames(WGS_ABSOLUTE_SP_sigs$solution_list$S4$Signature.norm) <- paste0("WGS_ABSOLUTE_sigs", seq(1:4))
colnames(WGS_aceseq_SP_sigs$Signature.norm) <- paste0("WGS_ACEseq_sigs", seq(1:4))
sim <- get_sig_similarity(WGS_ABSOLUTE_SP_sigs$solution_list$S4, WGS_aceseq_SP_sigs)p <- pheatmap::pheatmap(sim$similarity, cluster_cols = F, cluster_rows = F, display_numbers = TRUE)
Compare the data platforms
CNA signatures have been extracted independently from WGS, WES and SNP array derived prostate cancer CNA profiles. We compare the 4 copy number signatures extracted by Sigprofiler platform.
SNP vs WGS
SNP_sigs_source <- readRDS("../data/benchmark/SNP_ABSOLUTE_SP_source.rds")
WGS_sigs_source <- readRDS("../data/benchmark/WGS_ABSOLUTE_SP_source.rds")
WES_sigs_source <- readRDS("../data/benchmark/WES_FACETS_SP_source.rds")
colnames(SNP_sigs_source$solution_list$S4$Signature.norm) <- paste0("SNP_sigs", seq(1:4))
colnames(WGS_sigs_source$solution_list$S4$Signature.norm) <- paste0("WGS_sigs", seq(1:4))
colnames(WES_sigs_source$solution_list$S4$Signature.norm) <- paste0("WES_sigs", seq(1:4))
sim <- get_sig_similarity(SNP_sigs_source$solution_list$S4, WGS_sigs_source$solution_list$S4)p <- pheatmap::pheatmap(sim$similarity, cluster_cols = F, cluster_rows = F, display_numbers = TRUE)
SNP vs WES
sim <- get_sig_similarity(SNP_sigs_source$solution_list$S4, WES_sigs_source$solution_list$S4)
p <- pheatmap::pheatmap(
sim$similarity,
cluster_cols = F,
cluster_rows = F,
display_numbers = TRUE,
legend_breaks = c(0.2, 0.4, 0.6, 0.8)
)
### WES vs WGS
sim <- get_sig_similarity(WES_sigs_source$solution_list$S4, WGS_sigs_source$solution_list$S4)
p <- pheatmap::pheatmap(sim$similarity, cluster_cols = F, cluster_rows = F, display_numbers = TRUE)